ceil/floor for DatetimeIndex

[mayankanand007]

Follow-up to #9571 where we add ceil and floor support for Series.

Here we add ceil and floor support to DatetimeIndex class. This PR is dependent on #9571 getting merged first since it assumes the libcudf implementation for floor exists.

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (branch-22.02@3b38aa7). Click here to learn what that means.
The diff coverage is n/a.

@@               Coverage Diff               @@
##             branch-22.02    #9554   +/-   ##
===============================================
  Coverage                ?   10.60%           
===============================================
  Files                   ?      118           
  Lines                   ?    20090           
  Branches                ?        0           
===============================================
  Hits                    ?     2130           
  Misses                  ?    17960           
  Partials                ?        0           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 3b38aa7...b8ff1b4. Read the comment docs.

[mayankanand007]

Based on a slack conversation with Ashwin:

The Index class in cuDF derives from a Frame class, which is a common base class for Series, DataFrame and Index.
Frame is where ceil is currently implemented. That ceil implementation does not currently support a freq= argument.
We should:

1. Add a ceil (and floor) method in the subclasses DatetimeIndex and TimedeltaIndex that do support a freq argument.
2. Deprecate the Frame.ceil and Frame.floor methods.

[shwina]

but I wonder if we really need to deprecate the Frame methods.

My 2c is yes. I realize at this point that it's more effort to deprecate these APIs than just to continue supporting them. However, compatibility goes both ways: if users want to switch out Pandas with cuDF with minimal effort, doing the same the other way should also be minimal effort, and enabling APIs like these makes that more difficult.

[isVoid]

I don't know the history but I imagined back in the days when UDFs where not handy, these functions are very useful rounding columns. Today with improvements in UDF it can be done like df.apply(lambda row: math.ceil(row['a']), axis=1). I see why we don't want to maintain two approach for the same goal.

[mayankanand007]

@shwina should I then add Deprecation warning or no?

[shwina]

Yes please :)

[isVoid]

One small suggestion, else lgtm!

[shwina]

@gpucibot merge